[GitHub] [tvm] yzh119 commented on issue #15342: [Bug] The optimal implementation of reduce_sum searched by Ansor is more than 30x slower than torch.sum

["yzh119 (via GitHub)" <gi...@apache.org>]

yzh119 commented on issue #15342:
URL: https://github.com/apache/tvm/issues/15342#issuecomment-1641079322

   Hi @MrJungle1 , firstly, the previous CUDA codegen is suboptimal, some recent improvements such as multi-warp reduction (https://github.com/apache/tvm/pull/15327) should help.
   
   Besides, the way you profile PyTorch CUDA kernels is problematic because of synchronous execution, `time.time()` would return before the kernel execution finishes. please either insert `torch.cuda.synchronize()` barriers or uses PyTorch's native profiler.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org